Research map / Video / Video segmentation and tracking
Unique Lives, Shared World: Learning from Single-Life Videos
Tengda Han, Sayna Ebrahimi, Dilara Gokay, Li Yang Ku, Maks Ovsjanikov, Iva Babukova, Daniel Zoran, Viorica Patraucean, Joao Carreira, Andrew Zisserman, Dima Damen
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| train on egocentric single-life video instead of non-life control videos | CroCo single-life models · CAS with released CroCo | CAS: near-zero for non-life controls vs strong alignment for lives | Improved |
| increase single-life training duration to 30 min - 2 h (ablation) | CroCo single-life models · CAS with released CroCo | CAS: alignment emerges at 30 min (Walking Tours), 1 h (ALD), 2 h (HD-Epic) | Improved |
| train on 30h of one ALD life instead of 30h of K400 | CroCo · NYU-Depth-v2 and HPatches | depth accuracy / AEPE: ALD average nearly matches or surpasses K400 at 30h | No clear effect |
| train on 30h of single life instead of 30h of K400 | CroCo · ScanNet depth | AbsRel: K400 shows distinct advantage | Hurt |
| use union of spatial and temporal pairing (ablation) | CroCo single-life (HD-Epic) · ScanNet / NYU depth | relative gain over augmented pairing | Improved |
| use temporal pairing instead of augmented or random pairing (ablation) | CroCo single-life (Walking Tours) · ScanNet / NYU depth | relative gain | Improved |
| use random pairing instead of augmented pairing (ablation) | CroCo single-life (ALD) · ScanNet / NYU depth | relative gain | Improved |
| swap CroCo objective for DINOv2 objective | single-life Walking Tours models · NYU-Depth-v2 delta1 | accuracy: 0.538 -> 0.562 on NYU; HPatches AEPE 19.4 -> 21.2 (worse) | Mixed |
Closest papers by meaning
- Towards Stable Self-Supervised Object Representations in Unconstrained Egocentric Video (CVPR 2026 · similarity 0.87)
- Bootstrap Your Own Views: Masked Ego-Exo Modeling for Fine-grained View-invariant Video Representations (CVPR 2025 · similarity 0.85)
- MultiPly: Reconstruction of Multiple People from Monocular Video in the Wild (CVPR 2024 · similarity 0.84)
- GEOPHYS: The Geometry of Physical Plausibility (NeurIPS 2026 · similarity 0.84)
- Test-Time Adaptation for Combating Missing Modalities in Egocentric Videos (ICLR 2025 · similarity 0.84)
- Viewpoint Rosetta Stone: Unlocking Unpaired Ego-Exo Videos for View-invariant Representation Learning (CVPR 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
